Multi-class Generative Adversarial Nets for Semi-supervised Image Classification

TL;DR

This paper introduces a modified training method for GANs to improve multi-class semi-supervised image classification, especially when classes have similar features, demonstrated on MNIST datasets.

Contribution

It proposes a novel modification to traditional GAN training to enhance classification accuracy among similar classes in a semi-supervised setting.

Findings

Improved classification accuracy on MNIST and Fashion-MNIST datasets.

Demonstrated that traditional GANs struggle with similar classes.

Modified GAN training reduces class confusion in semi-supervised learning.

Abstract

From generating never-before-seen images to domain adaptation, applications of Generative Adversarial Networks (GANs) spread wide in the domain of vision and graphics problems. With the remarkable ability of GANs in learning the distribution and generating images of a particular class, they can be used for semi-supervised classification tasks. However, the problem is that if two classes of images share similar characteristics, the GAN might learn to generalize and hinder the classification of the two classes. In this paper, we use various images from MNIST and Fashion-MNIST datasets to illustrate how similar images cause the GAN to generalize, leading to the poor classification of images. We propose a modification to the traditional training of GANs that allows for improved multi-class classification in similar classes of images in a semi-supervised learning framework.